OpenAI is publicly diverging from the Trump administration on one of the most consequential questions facing the artificial intelligence industry: who should oversee the testing of advanced AI systems and whether those reviews should be mandatory before powerful models are released.
The debate emerged this week as OpenAI released a policy framework advocating for compulsory evaluations of frontier AI models, while the White House continues to promote a voluntary review system established under President Donald Trump's June 2 executive order.
The disagreement highlights a growing divide in Washington over how to balance innovation, national security, and government oversight as increasingly capable AI systems move from research labs into the broader economy.
OpenAI's proposal centers on requiring advanced AI models to undergo risk assessments before deployment. According to a policy paper discussed by company executives and reported by Politico, the company believes those evaluations should be conducted by the Center for AI Standards and Innovation (CAISI), a division of the National Institute of Standards and Technology within the Commerce Department.
The recommendation contrasts with the administration's newly announced framework, which allows developers to voluntarily provide federal officials with access to advanced models for cybersecurity testing. Under the White House approach, agencies including the National Security Agency would play a significant role in evaluating potential risks associated with advanced AI systems.
"You also have CAISI that has developed into something and has the capability to really do the type of sophisticated testing," Lehane said.
The White House executive order reflects growing concerns that advanced AI systems could eventually be exploited in cyberattacks, intelligence operations, or efforts targeting critical infrastructure. The administration directed federal agencies to strengthen cybersecurity protections, establish an AI cybersecurity clearinghouse, and develop classified benchmarking systems to assess advanced AI cyber capabilities.
National security concerns have become a central theme in the AI policy debate as governments increasingly examine how frontier models could affect sectors ranging from energy and finance to defense and communications. Administration officials have argued that maintaining U.S. leadership in artificial intelligence requires both innovation and stronger safeguards against misuse.
OpenAI agrees that advanced systems require greater scrutiny, but the company differs on how that scrutiny should be implemented. While the White House order explicitly rejects government licensing, permitting, or pre-clearance requirements for AI models, OpenAI is advocating for mandatory evaluations before deployment.
Lehane said the company does not believe developers should be left alone to determine whether a model warrants independent review. At the same time, OpenAI stopped short of endorsing broader government approval requirements for releasing new models, instead emphasizing the need for a credible testing framework.
